Level of aggregation of zoning and temporal transferability of the gravity distribution model: The case of Lyon

This study deals with the temporal transferability of the parameters of the gravity model of trip distribution and focuses on the trade-off between spatial resolution and data requirements. The models are calibrated using O–D matrices constructed from the three most recent Lyon household travel surveys (1985, 1995 and 2006) and generalised travel time data from coded transport networks for the three dates. Calibration has been conducted for three different zoning levels which have been chosen in line with common practice. The parameters obtained from model calibration are then applied to estimate O–D matrices at a later date and the results are compared using indicators that have been established for the zoning level applied in calibration, but also using indicators that have been aggregated in two different ways: aggregation to create larger zones or distance segments. Our findings confirm our initial intuition: the choice of zoning is fundamentally important. Moreover, in the best case, the parameters of the model change, but not sufficiently for the goodness-of-fit of the "predicted" model to be very different from that of the matrix obtained during calibration. It is possible to use the gravity model for forecasting purposes, but on condition that the goals of the study are compatible with the level of error in the reproduction of the observed matrices. If the zoning is either too coarse or too fine grained, forecasting performance is compromised.